Microscopic Slice Image Processing and Automatic Labeling - Biomedical Signal Analysis: Contemporary Methods and Applications - page 368

erable performance increase, especially for larger images. Furthermore,
this approach can be readily extended to the case of MLPs by replac-
ing each neuronal activity calculation with the FFT-perceptron weight
calculation [29].

Evaluation

After the confidence map has been generated, it can be evaluated by
simple maxima analysis. However, as seen in figure 13.8, due to noise
and non cell objects in the images, maxima do not always correspond to
cell positions, so thresholding in the confidence map has to be applied
first. Values of 0 . 5to0 . 8 yield good results in experiments, and if a
neural network-based approach is taken, the threshold values are already
implicitly given by the bias value at the output neuron. Furthermore,
the cell classifier yields high values corresponding to a single cell when
applied to image patches with large overlap. Therefore, after a maximum
has been detected, adjacent points in the confidence map are also set to
zero within a given radius (15 to 18 were good values for 20

×

20 image
patches). Iterative application of this algorithm then gives the final cell
positions, and hence the image segmentation and the cell count.

13.6 Relation to Other Methods

Although feature counting per se has not been studied very intensely
(see, e.g., [183, 184]), it can of course be interpreted as a secondary
problem in the larger field of image segmentation . Its goal is to de-
compose one or multiple images into their “natural” parts, those being
specified by similarities such as color, shape, texture, or some higher-
level semantic meaning. Our problem of cell counting can then be solved
by counting those image segments that represent cells - in the case of
a perfect segmentation, these should consist of all components except
for the large background component (which itself could contain multiple
segments).
Nowadays common algorithms for image segmentation, apart from
neural network based approaches like the above, include segmentation
using morphological operations or linear decomposition algorithms such
as non negative matrix factorization [150]. A very common technique
belonging to the first category is the so-called watershed transform .Its